AI-Enabled Wearable Ultrasound for Noninvasive Central Venous Pressure Monitoring

Area of Science:

• Critical Care Medicine
• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Central venous pressure (CVP) is vital for assessing right ventricular preload but invasive measurement via central venous catheters (CVCs) is challenging for continuous bedside monitoring.
• Current CVP monitoring methods are invasive, time-consuming, and not ideal for real-time bedside assessment in acute and critical care settings.

Purpose of the Study:

• To develop and clinically validate an artificial intelligence (AI)-enabled wearable ultrasound device for noninvasive estimation of elevated CVP.
• To assess the accuracy and feasibility of the AI device in critically ill patients compared to traditional CVC measurements.

Main Methods:

• A prospective, multi-center clinical study involving 349 intensive care unit (ICU) patients.
• Utilized a wearable ultrasound device for internal jugular vein (IJV) and common carotid artery (CCA) imaging.
• Employed a dual-decoder spatiotemporal attention network (DSTA-Net) for vascular segmentation and a dual-modality multilayer perceptron (DM-MLP) for CVP prediction.

Main Results:

• The DSTA-Net model demonstrated high accuracy and strong correlation with manual measurements for IJV and CCA quantification.
• The DM-MLP model achieved superior CVP prediction performance, with AUCs of 0.91 (internal) and 0.87 (external test sets).
• The AI-enabled device showed excellent agreement with invasive CVP measurements.

Conclusions:

• An AI-integrated wearable ultrasound device provides accurate, noninvasive bedside CVP assessment in acute and critical care patients.
• This technology presents a promising alternative to invasive CVCs for hemodynamic monitoring.
• The developed AI models (DSTA-Net and DM-MLP) significantly enhance the accuracy of noninvasive CVP estimation.